These were printed as raw bytes until 2026-09-22, with a note saying the share reading was wrong: they read 10000 in 3,874 of 3,880 single-holder configs, yet 0 of 120 multi-holder configs summed to 10000. The control was right to stop the claim and wrong about why. The field was always basis points. The stride was wrong. An entry is 34 bytes, not 40, so every holder after the first was being read six bytes inside the previous one’s share. Re-read at 34, 24 of 24 multi-holder configs sum to exactly 10000.